Naval Ravikant, co-founder of AngelList, discusses his deep dive into 'vibe coding' with AI agents following the December 2025 release of Claude Opus 4.5. Despite not having coded seriously in decades, Naval has become addicted to building custom apps using conversational AI that connects directly to Unix systems and can execute commands, manage files, and spawn tasks.

The conversation covers Naval's creation of a personal app store that delivers one-shot custom applications to his iPhone, his prediction that pure software startups are becoming uninvestable, and his belief that Apple's dependence on external AI models threatens the company's long-term dominance. Naval demonstrates how AI coding agents can handle everything from bug fixes to feature development, potentially enabling single-person software companies to achieve massive scale.

The Claude Opus 4.5 Inflection Point for AI Coding

"Around December of 2025, the coding agents in AI hit an inflection point with the release of Claude Opus 4.5" - Naval describes agents that "stay on track, can build apps soup to nuts, can solve thorny problems, and really feels like having a junior programmer at your disposal who's fast, essentially free, and ready to please."

These agents operate through command line interfaces connected to Unix systems, leveraging their training on GitHub code and Stack Overflow to understand Unix commands like grep, awk, sed, and pipe operations.

"AIs are incredible translators" - they convert between programming languages (Python, C, Lisp, Rust) and English, making coding accessible without knowing specialized tools and commands.

Building a Personal App Store for One-Shot Applications

Naval created a personal app store where he can request custom apps and receive them instantly: "I can ask it for an app. It can deliver that app to my app store, which is a web page. And eventually I made it into an app itself that lives on my iPhone."

Example workout app creation: Naval requested an app using "functionality of tonal and ladder, follow Apple's human interface guidelines" with custom tracking, graphs, strength scores, body diagrams, and Apple Health integration - delivered as a working app immediately.

"You can literally be at dinner with someone, having a conversation. They describe some app they want. You can describe it to Claude. And five minutes later, you're showing them that app on your phone."

The End of Venture-Investable Pure Software

"Pure software is uninvestable" - Naval argues that if your advantage is building software others can't, "that's uninvestable" because anyone can now hack it together and coding agents improve rapidly.

Venture investors must now focus on "hardware, network effects, AI models" with training AI models becoming "the new building software for however long that lasts until auto research and auto training starts working."

The shift enables individual creators: "There's never been a better time to be alive as a creator of software" though scaling to market still requires traditional engineering teams.

Automated Bug Fixing and Development Workflows

Naval implemented automated bug resolution: "I have Claude go every 24 hours through all the bug reports and it just fixes them all by itself without my having to intervene."

The system creates fixes in side branches for human review: "All I have to do is just review the fixes and say, oh, that wasn't really a bug. That wasn't a good fix. Don't ship that."

Multiple AI models provide code review through automated pull requests: "Codex and Gemini automatically fire in every pull request" creating "a roundtable of AIs" for collaborative development.

Apple's Strategic AI Mistake and iPhone Disruption

"Apple giving up on AI will go down as the biggest strategic mistake in the tech industry of this decade" - Naval predicts this caps Apple's growth and threatens its dominance.

When users interact primarily with AI agents instead of apps ("call me an Uber" vs opening Uber app), "the need for a phone becomes much smaller" as agents create interfaces on-demand.

Apple's reliance on Google's Gemini undermines differentiation: "Apple using Gemini, which is Google's AI model. So, what's the difference? I might as well just use an Android phone."

Context Window Limitations and Model Management

Current models handle ~1 million tokens but struggle with complex codebases: "The context window runs out as your code base gets larger. The models can't keep all of it in memory anymore."

Models make architectural mistakes when context is exceeded: "They start fixing the wrong thing, they fix the same bug five times, they go do a quick patch in the architecture when the problem lies somewhere else."

Human oversight remains critical: "You do have to guide it. It does require a lot of operational oversight" as models are "always trying to please you" like "a dog" that's better at specific tasks but needs direction.

Why AI Excels at Coding vs Other Domains

Coding provides ideal training conditions: "There's tons and tons of data" and "the code has to compile and has to execute" with "simple tests that are pre-written" for verification.

Mathematics and self-driving benefit from similar advantages: "You have a ton of data, you have a lot of solved problems, and you can verify the output very easily."

Creative domains lack verification mechanisms: "In creative writing, like who determines what's good creative writing versus what's not" making it harder to train models effectively.